Open urlicht opened 11 months ago
Yes, I was absolutely unable to register a package update on GitHub/Registrator because of this. My solution was install matplolib through Conda.pip and then Conda.add, before pointing PyCall to Conda's python. It works now, see here for my solution within a build script. This needs to remain an open issue IMHO.
Here is a reproducer using docker:
docker run -it julia:1.9 /bin/bash
julia --eval 'using Pkg; Pkg.add("PyPlot");'
julia --eval 'using PyPlot'
I am wondering if it also affects non-Linux systems.
Other workarounds:
As it is likely a conda problem, I posted this question of StackOverflow: https://stackoverflow.com/questions/76817732/installing-matplotlib-with-conda-fails-with-conda-exceptions-resolvepackagenotfo
Not sure if it's a PyPlot-specific problem (could be caused by
PyCall.jl
ormatplotlib
itself I suppose)Upon a fresh install of the latest release of PyCall (setting
ENV["PYTHON"]=""
to use a separate Julia miniconda environment), it fails to install certain packages, including matplotlib. Trying to installPyPlot.jl
(or even manually activating the environment and performingconda install matplotlib
) returns the following error:Reproducible across multiple Ubuntu machines running 18.04, 20.04, or 22.04